Mining Association Patterns From Neighborhood Insight.

Area of Science:

• Data Mining
• Statistical Association Analysis
• Granular Computing

Background:

• Detecting complex association patterns between variables is crucial but challenging due to data complexities.
• Existing association measures often lack generality or equitability, failing to capture diverse structures without bias.
• Granular computing, utilizing local neighborhood structures, offers a promising approach for multi-scale association information.

Purpose of the Study:

• To introduce a novel association measure, the maximal neighborhood coefficient (MNC), that addresses limitations of existing methods.
• To develop a family of maximal neighborhood nonparametric exploration (MNNE) statistics for richer association characterization.
• To provide a data-driven toolkit for exploring complex association patterns with improved performance.

Main Methods:

• Developed the maximal neighborhood coefficient (MNC) based on k-nearest neighbors (k-NN) granulation.
• Introduced maximal neighborhood nonparametric exploration (MNNE) statistics as extensions to MNC.
• Employed granular computing principles to capture multi-scale association information.

Main Results:

• MNC demonstrates the ability to capture a broad range of associations without empirical bias.
• MNC retains local structural details often missed by traditional association measures.
• MNNE statistics provide richer auxiliary information for characterizing complex associations.

Conclusions:

• MNC and MNNE form a powerful toolkit for data-driven exploration of complex association patterns.
• The proposed methods offer a new perspective on mining associations, overcoming limitations of existing techniques.
• The toolkit exhibits strong empirical performance in identifying diverse and unbiased association structures.